A Cascaded Unsupervised-Supervised NLP Pipeline for Detecting Accusatory Language in Public Procurement

Bryan Torres, Daniel Riofrío, José Vega-Sánchez, Nathaly Orozco, Carla Parra et al.|August 12, 2026arXiv

Key Takeaway

Lightweight, domain-specific NLP models can effectively detect fraud signals in real-world government data without expensive infrastructure—useful for building practical oversight tools in resource-constrained settings.

Summary

This paper develops a hybrid NLP pipeline to detect accusatory language in public procurement comments from Ecuador's procurement system.

applications data

Key Terms

word2vec gaussian-mixture-model semantic-embeddings class-imbalance